Poster Session P9.1 Tornado identification using a neuro-fuzzy approach to integrate shear and spectral signatures

Wednesday, 8 November 2006
Pre-Convene Space (Adam's Mark Hotel)
Yadong Wang, Univ. of Oklahoma, Norman, OK; and T. Y. Yu, M. Yeary, A. M. Shapiro, S. Nemati, M. Foster, and D. L. Andra Jr.

Handout (168.8 kB)

Current tornado detection algorithm (TDA) searches strong and localized azimuth shears in the field of mean Doppler velocity. Recent studies have shown that wide and flat spectra can often be observed in the region of tornado vortex by the research Weather Surveillance Radar-1998 Doppler (WSR-88D). The tornado spectra are significantly different from the spectra from other regions and therefore have the potential to facilitate tornado identification. The tornado spectral signature (TSS) can be generally characterized by large spectrum width. However, spectrum width depends on several other factors such as shear and the signal-to-noise ratio (SNR), for example. In addition, spectrum width sometimes does not provide sufficient shape information of the spectrum. Therefore, other three parameters are developed to characterize TSS. The dependence of the four parameters on range is studied statistically using simulations for various conditions such as the presence of shear and mesocyclone and different reflectivity structures.

A neuro-fuzzy system is developed in this work to integrate the TSS and the shear signature with the goal of improving tornado detection. The neuro-fuzzy approach was originally designed for the cases when the input information is vague and the classes are not disjoint. For example, in tornado detection a wide spectrum can be produced by tornadoes, strong shears, or low SNR. A decision based on the threshold of a single parameter like spectrum width could lead to false detection. On the other hand, a neuro-fuzzy system can avoid the threshold and integrate all the available information. Moreover, the neuro-fuzzy system can be optimized through a self-learning process. In this work, the neuro-fuzzy tornado detection algorithm (NFTDA) is demonstrated using tornadic data collected by the research WSR-88D (KOUN) in Norman, Oklahoma. KOUN is operated by the National Severe Storms Laboratory (NSSL) and has unique capability of continuously collecting Level I time series data. This data is described as 32 I&Q samples per range gate in our experiments,which leads to the spectral calculations for each range gate. Based on the 10 May 2003 tornado case in Edmond, Oklahoma, our NFTDA results are consistent with the tornado damage path. NFTDA also shows improved detection results compared to the conventional shear-based TDA. Furthermore, the impact of radar angular sampling on NFTDA is also investigated and discussed.

- Indicates paper has been withdrawn from meeting
- Indicates an Award Winner